Piezoelectric nanoelectromechanical resonators based on aluminum
nitride thin films
R. B. Karabalin,
1
M. H. Matheny,
1
X. L. Feng,
1
E. Defaÿ,
2
G. Le Rhun,
2
C. Marcoux,
2
S. Hentz,
2
P. Andreucci,
2
and M. L. Roukes
1,
a
1
Kavli Nanoscience Institute, California Institute of Technology, Pasadena, California 91125, USA
2
CEA LETI MINATEC, F-38054 Grenoble Cedex, France
Received 22 June 2009; accepted 23 July 2009; published online 9 September 2009
We demonstrate piezoelectrically actuated, electrically tunable nanomechanical resonators based on
multilayers containing a 100-nm-thin aluminum nitride
AlN
layer. Efficient piezoelectric actuation
of very high frequency fundamental flexural modes up to
80 MHz is demonstrated at room
temperature. Thermomechanical fluctuations of AlN cantilevers measured by optical interferometry
enable calibration of the transduction responsivity and displacement sensitivities of the resonators.
Measurements and analyses show that the 100 nm AlN layer employed has an excellent
piezoelectric coefficient,
d
31
=2.4 pm
/
V. Doubly clamped AlN beams exhibit significant frequency
tuning behavior with applied dc voltage. ©
2009 American Institute of Physics
.
doi:
10.1063/1.3216586
Piezoelectricity, a direct electrical-mechanical conver-
sion occurring in certain types of crystals, has found numer-
ous applications ever since its discovery by the Curie broth-
ers in 1880.
1
Some of the most ubiquitous among these are
electrical oscillators, clocks, and microbalances based on pi-
ezoelectric crystals such as quartz.
2
,
3
Recent development of
micromachining techniques has greatly facilitated the real-
ization of various miniaturized piezoelectric devices, includ-
ing zinc-oxide
ZnO
beam-structured,
4
disk,
5
and film bulk
acoustic resonators
FBARs
,
6
the noteworthy Agilent’s com-
mercialized AlN FBARs,
7
AlN contour-mode resonators and
filters,
8
and monolithically integrated FBAR-complementary
metal oxide semiconductor
CMOS
filters.
9
All these piezo-
electric microelectromechanical systems
MEMS
have re-
lied on active layers of approximately micron scale thickness
that provide strong electromechanical coupling.
The emerging field of nanoelectromechanical systems
NEMS
is attracting considerable interest. These miniatur-
ized nanoscale devices, particularly cantilever and beam
flexural-mode resonators, have enabled the demonstrations
of single-molecule mass sensors
10
and single-cell-level force
sensors.
11
For NEMS resonators in such applications, piezo-
electric actuation appears to be particularly advantageous
compared to the more conventionally employed
magnetomotive,
10
electrostatic,
12
and electrothermal
13
excita-
tions. Among its attributes are intrinsic integrability, high
efficiency and electrical tunability, low power consumption,
and low thermal budgets for materials processing permitting
post-CMOS integration.
5
–
9
Previously we have prototyped
piezoelectric NEMS using epitaxial gallium arsenide
GaAs
heterostructures.
14
To date, however, nanoscale resonators
have not yet been realized with materials having higher pi-
ezoelectric coupling efficiency. Here we employ AlN to ac-
complish this, since it has superior material properties such
as high acoustic velocity,
E
Y
/
, and low dielectric loss.
15
Among the principal challenges in its use for NEMS are
obtaining nanometer-thick AlN films that retain excellent pi-
ezoelectric properties and devising efficient actuation and
readout schemes. In this letter we demonstrate piezoelectric
NEMS resonators actuated by 100-nm-thick AlN layers, and
characterize their driven resonance response, noise spectra,
and frequency tuning.
The 100 nm AlN piezolayer is sandwiched between two
100 nm molybdenum
Mo
electrode layers. In our 200 mm
wafer-scale process, Mo films are prepared by dc magnetron
sputtering, and AlN film is sputtered by a dc pulsed magne-
tron reactive process. Prior to sputtering the bottom Mo film,
a 20 nm AlN seed layer on a
100
silicon
Si
substrate is
deposited. A 99.999% pure Aluminum
Al
disk is used as
the target in
3 mTorr of dry nitrogen
N
2
during sputter-
ing, carried out with 1500 W of applied power. This process
has recently been optimized
16
to produce highly
c
-axis tex-
tured AlN films. Thin film tests indicate that piezoelectric
coefficients close to those of bulk AlN can be preserved
down to
250–500 nm films.
17
Figure
1
displays a four-mask surface nanomachining
process we have developed to fabricate suspended cantile-
vers and beams using the Mo/AlN/Mo/seed-AlN
100 nm/
100 nm/100 nm/20 nm
stack. A key requirement for piezo-
electric actuation is to make electrical contacts to both top
and bottom Mo electrodes, as illustrated in Fig.
2
a
. In the
first step shown in Fig.
1
a
, a region is patterned by
electron-beam lithography
EBL
using spin-on-glass resist
resulting in a 170 nm silicon dioxide
SiO
2
“mesa” layer,
which protects the structural stack during subsequent pro-
cesses. Outside of the mesas, argon/chlorine
Ar
/
Cl
2
based
inductive coupled plasma
ICP
etch anisotropically removes
the top two layers and is terminated at the bottom Mo
Fig.
1
b
. Gold
Au
bonding pads are then patterned and depos-
ited on bottom Mo layer adjacent to mesa regions
Fig.
1
c
.
After parts of the bottom Mo are protected by an insulating
SiO
2
bridge, the remaining is removed by a wet etch, yield-
ing the device profile shown in Fig.
1
c
, with bottom contact
electrode displayed on the left and top electrode on the right.
This separate patterning of both top and bottom bonding
pads on the same Mo layer substantially reduces parasitics.
After stripping the SiO
2
, the mesa’s top Mo layer is ex-
posed and another SiO
2
bridge with conducting path on top
a
Electronic mail: roukes@caltech.edu.
APPLIED PHYSICS LETTERS
95
, 103111
2009
0003-6951/2009/95
10
/103111/3/$25.00
© 2009 American Institute of Physics
95
, 103111-1
Downloaded 25 Sep 2009 to 131.215.220.165. Redistribution subject to AIP license or copyright; see http://apl.aip.org/apl/copyright.jsp
is patterned between the mesa and top electrode
Fig.
1
d
.
As shown in Fig.
1
e
, the nanodevices are then patterned
with high resolution EBL followed by deposition of an insu-
lating strontium fluoride
SrF
2
mask. An anisotropic ICP
etch is used to transfer the pattern through all four structural
layers, down to the Si substrate. Finally, devices are sus-
pended by isotropically etching the Si sacrificial layer in an
argon/nitrogen trifluoride
Ar
/
NF
3
plasma. The scanning
electron micrograph
SEM
of a typical device
Fig.
2
b
shows the composing layers in the structure.
When a voltage signal is applied between the device’s
top and bottom electrodes
Fig.
2
a
, the piezoelectricity
tensor causes longitudinal strain to develop in the active
layer according to
s
xx
=
d
31
E
z
, where
s
xx
is the piezoelectric
longitudinal strain,
E
z
is a vertical component of electric
field, and
d
31
is the piezoelectric coefficient. The seed layer
offsets the neutral plane from the active layer’s central plane;
as a result, when the piezoelectric layer is strained, it causes
a bending moment in the structure. Upon application of high-
frequency actuation, devices are driven into resonance and
their response is characterized via two-port network analysis.
In our specific transduction scheme, the out-of-plane funda-
mental flexural-mode resonances are usually most promi-
nent. In order to avoid crosstalk between actuation and de-
tection signals we use an optical interferometric technique
18
to readout the mechanical displacement
Fig.
2
c
. All mea-
surements are performed at room temperature at a base pres-
sure of
5 mTorr. Figure
2
d
shows the measured reso-
nance of a typical AlN cantilever
Fig.
2
b
with dimensions
of
L
w
t
=6
m
900 nm
320 nm. A Lorentz fit to
the measured response yields a quality factor
Q
=960 for the
fundamental resonance at 9.11 MHz. The observed reso-
nance frequency closely matches our analytical prediction
9.05 MHz
for the four-layered cantilever, given by
19
f
0,
th
=
1.875
2
2
L
2
i
=1
4
E
Yi
w
i
n
=1
i
−1
t
n
n
=1
i
t
n
z
−
z
c
2
dz
/
i
=1
4
i
w
i
t
i
1
/
2
,
1a
where
z
c
=
i
=1
4
E
Yi
n
=1
i
−1
t
n
n
=1
i
t
n
zdz
/
i
=1
4
E
Yi
t
i
.
1b
Here
z
c
is the position of the structure’s elastic neutral plane
z
=0 is at the bottom of the multilayer
. The indices corre-
spond to four consecutive layers of the material:
1
seed
AlN, thickness
t
1
=20 nm, Young’s modulus
E
Y
1
=345 GPa;
2
bottom Mo,
t
2
=100 nm,
E
Y
2
=329 GPa;
3
active AlN,
t
3
=100 nm; and
4
top Mo with
t
4
=100 nm.
The displacement sensitivity of our readout scheme is
sufficient to detect the device’s thermomechanical noise. We
obtain the noise spectra of the undriven devices using a spec-
trum analyzer. Figure
2
e
demonstrates the thermomechani-
cal noise spectrum of the 9.11 MHz cantilever. In the limit of
Q
1, the thermomechanical displacement noise spectral
density on resonance is
S
z
,thm
1
/
2
=
k
B
TQ
/
2
3
f
0
3
M
eff
1
/
2
, where
k
B
,
T
, and
M
eff
are Boltzmann’s constant, temperature, and
the device’s effective mass, respectively. This yields the
room-temperature displacement noise spectral density on
resonance,
S
z
,thm
1
/
2
0.16 pm
/
Hz
1
/
2
. The spectrum in Fig.
2
e
has a background noise floor
S
v
,sys
1
/
2
90 nV
/
Hz
1
/
2
near the
resonance. The thermomechanical displacement noise den-
sity is transduced into a voltage noise density
S
v
,thm
1
/
2
=
S
v
,total
−
S
v
,sys
1
/
2
48 nV
/
Hz
1
/
2
. From this analysis we de-
termine a transduction responsivity
R
S
v
,thm
1
/
2
/
S
z
,thm
1
/
2
300 nV
/
pm, and a mechanical displacement sensitivity
S
z
,sys
1
/
2
=
S
v
,sys
1
/
2
/
R
0.3 pm
/
Hz
1
/
2
for the measurement system.
Mo
AlN
Si
Mo
AlN
SiO
2
Mo
Mo
AlN
Si
Mo
Au
Au
SiO
2
AlN
AlN
Mo
AlN
Si
Mo
AlN
SiO
2
Mo
AlN
Si
Mo
AlN
Au
SiO
2
Au
Mo
Mo
AlN
Si
Mo
AlN
Au
Au
SiO
2
Au
Mo
Au
Mo
AlN
Si
Mo
AlN
Au
SiO
2
Au
SrF
2
Mo
Au
Mo
AlN
Mo
SiO
2
Si
Au SrF
2
(
a
)
(b)
(c)
(d)
(e)
(f)
SrF
2
SrF
2
FIG. 1.
Color online
Surface nanomachining process flow for piezoelectric
AlN NEMS.
a
Definition of SiO
2
mesa, followed by a dry etch
b
to
remove top Mo and AlN.
c
Au electrodes are patterned and deposited
adjacent to the mesa region, along with an SiO
2
bridge protecting part of
bottom Mo, while the remaining bottom Mo was removed.
d
SiO
2
is
stripped and contact is made to the top Mo electrode
including deposition
ofaSiO
2
bridge layer
.
e
NEMS devices are defined, using SrF
2
as a dry
etch mask for both NEMS and all metallic contacts.
f
Anisotropic etching
of all the structural layers down to Si by using Ar
/
Cl
2
ICP-RIE is followed
by an isotropic Ar
/
NF
3
etch for device release, and subsequent removal of
SrF
2
mask. The dashed arrows and lightly hatched areas indicate that the
bottom AlN/Mo layers are connected out of the present section plane.
He/Ne
Laser
PhotoDetector
~
Vib ti
(c)
k
r
Lens
Splitter
(PD)
2
(a)
E
Field
Vib
ra
ti
on
Networ
k
Analyze
DCBlock
Lens
1
V
DC
Vacuum
200
1μm
(b)
100
110
032
0.34
0.36
D
(pm/Hz
1/2
)
n
V/Hz
1/2
)
150
0.4
0.6
MeasuredDrivenResponse
Data
Fit
Q
=960
n
t (nm)
(
μ
V)
2
/
1
thm
S
90
0.28
0.30
0
.
32
s
placementS
D
V
oltageSD (
n
50
100
0.2
Displaceme
n
Signal
2
/
1
S
9.08 9.10 9.12
80
0.26
Di
s
V
Frequency (MHz)
9.05 9.10 9.15
0
0.0
Frequency (MHz)
2
/
1
sys
S
(d)
(e)
FIG. 2.
Color online
Characterization of high frequency AlN cantilever
NEMS.
a
Illustration of piezoelectric actuation.
b
An SEM image dis-
playing the four-layer composite structure of a cantilever NEMS.
c
Sim-
plified schematic of the optical readout scheme
note for the undriven noise
measurements, port 1 is disconnected and network analyzer is replaced by a
spectrum analyzer; also
V
dc
is applied only during the tuning measurements
shown in Fig.
3
.
d
A typical resonant response measured from a driven
cantilever
device shown in
b
, with
f
0
=9.11 MHz and
Q
=960, extracted
from the fit to the model of damped driven oscillator.
e
Measured thermo-
mechanical noise spectral density from the 9.11 MHz device, demonstrating
the sensitivities in both electrical and mechanical domains.
103111-2
Karabalin
etal.
Appl. Phys. Lett.
95
, 103111
2009
Downloaded 25 Sep 2009 to 131.215.220.165. Redistribution subject to AIP license or copyright; see http://apl.aip.org/apl/copyright.jsp
With this calibration of the transduction responsivity, we
map out device’s motions in the mechanical domain.
For the response plotted in Fig.
2
d
, the displacement on
resonance is 0.54 nm at 2 mV drive. Following a detailed
numerical analysis of piezoelectrically deflected multilayer
cantilevers,
20
we determine the piezoelectric coefficient of
the 100 nm AlN layer to be
d
31
=2.4 pm
/
V. To complement
these measurements, we have also fabricated centimeter-
scale samples from the same wafer, and independently
measured the piezoelectric constant of the AlN layer by
means of a setup detailed elsewhere.
21
Such measurements
yield
d
31
=2.6 pm
/
V, a value consistent with that deter-
mined by the NEMS resonance measurements. These values
2.5 pm
/
V
are only slightly lower than those reported for
single crystals.
17
This demonstrates the high performance
that can be attained with 100 nm polycrystalline AlN films.
We further demonstrate operation of tunable, doubly
clamped beam AlN resonators. In beams such as displayed in
the SEM inset of Fig.
3
a
, the piezoelectric effect not only
generates actuation but also produces resonance frequency
tuning by application of a dc voltage. This particular device
has dimensions
L
w
t
=4
m
900 nm
320 nm, reso-
nance frequency
f
0
=78.2 MHz, and quality factor
Q
=670.
The moderate
Q
reduction
compared to the 9.11 MHz can-
tilever
for this smaller device with larger surface to volume
ratio is consistent with earlier observations
22
of a
Q
depen-
dence on the size of MEMS/NEMS resonators, likely due to
relatively larger clamping and surface losses. Resonance
curves measured with increasing drive amplitudes
Fig.
3
a
show that with modest excitation voltages, motion beyond
the onset of nonlinearity
23
is obtained. These measurements
agree with analytic predictions of the Duffing nonlinearity,
which arises from induced tension in doubly clamped beam
resonators.
24
As a dc voltage is applied between the top and bottom
electrodes, an electrical field is created across the AlN layer,
and the piezoelectrically induced strain is converted into a dc
longitudinal stress, subject to the clamped-clamped boundary
conditions. This modulation of the tensile stress leads to a
shift in resonance frequency. This behavior is verified by
monitoring the resonance frequency of a beam while
sweeping the dc voltage. Figure
3
b
shows the measured
frequency tuning as a function of the applied dc voltage. As
expected, the resonance frequency varies approximately lin-
early with the voltage. The measured frequency tunability of
34 kHz/V is consistent with results from finite element simu-
lations, plotted in Fig.
3
b
, where multiple layers and finite
ledges
1
m in width, resulting from the undercut associ-
ated with device release
are taken into account.
We demonstrate piezoelectric very high frequency tun-
able NEMS resonators, which offer significant potential for a
broad spectrum of applications.
We are grateful to G. Villanueva, J.L. Arlett, and J.E.
Sader for helpful discussions and Y. Wu for illustration. We
acknowledge financial support for this work from DARPA/
MTO and SPAWAR under the Grant No. N66001-07-1-2039.
1
P. Curie and J. Curie, Bull. Soc. Fr. Mineral.
3
,90
1880
.
2
W. G. Cady,
Piezoelectricity: An Introduction to the Theory and Applica-
tions of Electro-Mechanical Phenomena in Crystals
Dover, New York,
1964
.
3
Applications of Piezoelectric Quartz Crystal Microbalances
, edited by C.
Lu and A. W. Czanderna
Elsevier, New York, 1984
.
4
D. L. DeVoe,
Sens. Actuators, A
88
,263
2001
.
5
L. Yan, W. Pang, E. S. Kim, and W. C. Tang,
Appl. Phys. Lett.
87
, 154103
2005
.
6
H. Zhang and E. S. Kim,
J. Microelectromech. Syst.
14
, 699
2005
.
7
R. C. Ruby, P. Bradley, Y. Oshmyanksy, A. Chien, and J. D. Larson III,
Proc.-IEEE Ultrason. Symp.
1
, 813
2001
.
8
G. Piazza, P. J. Stephanou, and A. P. Pisano, J. Microelectromech. Syst.
16
, 319
2007
.
9
M.-A. Dubois, J.-F. Carpentier, P. Vincent, C. Billard, G. Parat, C. Muller,
P. Ancey, and P. Conti,
IEEE J. Solid-State Circuits
41
,7
2006
.
10
Y. T. Yang, C. Callegari, X. L. Feng, K. L. Ekinci, and M. L. Roukes,
Nano Lett.
6
, 583
2006
.
11
J. L. Arlett, J. R. Maloney, B. Gudlewski, M. Muluneh, and M. L. Roukes,
Nano Lett.
6
, 1000
2006
.
12
R. R. He, X. L. Feng, M. L. Roukes, and P. D. Yang,
Nano Lett.
8
, 1756
2008
.
13
I. Bargatin, I. Kozinsky, and M. L. Roukes,
Appl. Phys. Lett.
90
, 093116
2007
.
14
S. C. Masmanidis, R. B. Karabalin, I. De Vlaminck, G. Borghs, M. R.
Freeman, and M. L. Roukes,
Science
317
, 780
2007
.
15
K. Tsubouchi, K. Sugai, and N. Mikoshiba, Proc.-IEEE Ultrason. Symp.
1981
, 375.
16
F. Martin, P. Muralt, and M. A. Dubois,
J. Vac. Sci. Technol. A
24
,946
2006
.
17
F. Martin, P. Muralt, M. A. Dubois, and A. Pezous,
J. Vac. Sci. Technol. A
22
, 361
2004
.
18
D. W. Carr and H. G. Craighead,
J. Vac. Sci. Technol. B
15
, 2760
1997
.
19
R. J. Roark and W. C. Young,
Formulas for Stress and Strain
McGraw-
Hill, New York, 1975
.
20
J. G. Smits and W. S. Choi,
IEEE Trans. Ultrason. Ferroelectr. Freq. Con-
trol
38
, 256
1991
.
21
E. Defay, C. Zinck, C. Malhaire, N. Baboux, and D. Barbier,
Rev. Sci.
Instrum.
77
, 103903
2006
.
22
P. Mohanty, D. A. Harrington, K. L. Ekinci, Y. T. Yang, M. J. Murphy, and
M. L. Roukes,
Phys. Rev. B
66
, 085416
2002
.
23
A. H. Nayfeh and D. T. Mook,
Nonlinear Oscillations
Wiley, New York,
1979
.
24
H. W. C. Postma, I. Kozinsky, A. Husain, and M. L. Roukes,
Appl. Phys.
Lett.
86
, 223105
2005
.
2.0
1.5
15
20
(mV)
nt (nm)
1μm
0.5
1.0
5
10
Signal
D
isplaceme
n
77.8 78.0 78.2 78.4 78.6
0.0
0
5
D
(a)
Frequency (MHz)
50
100
MeasuredData
Simulation
h
ift (kHz)
(a)
3
2
1
0
1
2
3
-100
-50
0
r
equencyS
h
-
3
-
2
-
1
0
1
2
3
F
r
DCVoltage (V)
(b)
FIG. 3.
Color online
VHF AlN beam resonators demonstrating nonlinear-
ity and frequency tuning behavior.
a
The measured 78.2 MHz resonance
clearly manifests the Duffing nonlinearity as the excitation is increased.
Inset is an SEM image of the device employed.
b
Measured and computed
piezoelectric frequency tuning as a function of dc polarization voltage
across the sandwiched AlN layer. Insets illustrate the static longitudinal
strain distribution
upper right
induced by a 2.5 V dc polarization, and the
mode shape
lower left
corresponding to the measured resonance.
103111-3
Karabalin
etal.
Appl. Phys. Lett.
95
, 103111
2009
Downloaded 25 Sep 2009 to 131.215.220.165. Redistribution subject to AIP license or copyright; see http://apl.aip.org/apl/copyright.jsp